Hi Ben,

On Mon, Jun 27, 2011 at 1:43 PM, Ben Niven-Jenkins
<[email protected]> wrote:
> Vijay,
>
> On 27 Jun 2011, at 17:23, Vijay K. Gurbani wrote:
>
>> As individual ...
>>
>> In reference to [1], where we discuss whether or not there
>> should be a 1-1 relationship between an IP address and
>> a PID, and if it is not, how should we handle it?
>>
>> On 06/24/2011 10:46 AM, Ben Niven-Jenkins wrote:
>>> I think we should say something along the lines of what I suggest (it
>>> probably doesn't need to go in much, if any, more detail either)
>>> otherwise we risk interoperability issues where one client may
>>> interpret such a map as invalid while others do not.
>>
>> Ben: The bigger question to me is: is it indeed valid to have
>> a 1:N mapping between an IP address and a PID?  If it is not,
>> then we leave this to the category of configuration errors and
>> not discuss how to handle it in the draft (the draft cannot
>> provide hedges against every MUST and MUST NOT).
>
> I have no issue with the current requirement that for a given IP address a 
> client should be able to resolve one (and only one) PID via longest prefix 
> matching and I'm not suggesting we change that.
>
> However I do think that a sentence or two on what it means when such a 
> scenario happens is useful for interoperability.
>
> For example, if my client uses the network map service to obtain the complete 
> network map and discovers that a particular IP address exists in two PIDs 
> (i.e. a longest prefix match would return two PIDs), how is my client 
> supposed to interpret the network map:
>
> a) Does it mean the entire network map is invalid and I should throw it all 
> away?
> b) Does it mean those PIDs containing the duplicate IP address are invalid 
> and I must disregard their complete contents but the rest of the map is 
> useable?
> c) Does it mean that those PIDs containing the duplicate IP address are valid 
> (along with the rest of the network map) provided I disregard the duplicate 
> IP addresses?
> d) Does it mean I am allowed to make a local decision as to which PID is the 
> correct PID for that IP address?
>
> IMO (a) & (b) require a client to fully validate the network map looking for 
> duplicate IP addresses before using any portion of it can be considered 
> useable. (c) & (d) allow a client to discover & recover from an inconsistency 
> at "run time".
>
> Validating the entire network map before using it may not have a significant 
> performance impact for an ALTO client in some P2P software where the number 
> of PIDs and IP prefixes in the network map is likely to be small, however for 
> some CDN use cases it could well be the case that there are thousands of PIDs 
> and millions of individual prefixes in a single network map. A simple example 
> would be if one wanted to represent in an ALTO network map a geo-location 
> database containing the entire global routing table.
>

If I were implementing a client, I would choose (d) in the interest of
Postel's Law.  I would also send an email to the ALTO Server's
provider or implementer notifying them that it is not compliant and
that they should stop it :)

Saying something in the protocol is okay with me, but a concern is
getting into the habit of trying to find and suggest possible
behaviors in case any of the MUST / MUST NOT's in the document are
violated.

Rich

> Ben
>
> _______________________________________________
> alto mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/alto
>
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto

Reply via email to